Nonlinear Model Predictive Control for Self-Driving cars Trajectory Tracking in GNSS-denied environments

In this paper, a trajectory tracking control problem for a self-driving vehicle is addressed using a nonlinear model predictive controller (NMPC). The solution relies on a direct multiple shooting method for discretizing the simplified differentially flat bicycle model which was used as a prediction plant to reduce the computational cost of NMPC. The solution makes use of the long prediction horizon of the NMPC, which allows safe trajectory tracking and achieving a user-specified goal location by controlling the longitudinal velocity and the steering angle. GNSS-denied navigation algorithm of the vehicle employs Visual Inertial Odometry (VIO) to obtain the position and orientation of car without using GNSS data. Simulation experiments is conducted to show the effectiveness of the system in tracking desired trajectories in high precision with speeds up to 1 m/s.